Real-time traffic monitoring and forecast through OPTIMA

- Optimal Path Travel Information for Mobility Actions

Lorenzo Meschini1, Guido Gentile1

Abstract

OPTIMA is an innovative ITS (Intelligent Transport Systems) platform for traf-

fic monitoring and management, that provides off-line estimates and real-time fore-
casts concerning the use (vehicle flows) and the performances (travel times) of the
road networks.
To this end, OPTIMA exploits an advanced methodology for modelling transport
demand and vehicle congestion, through which the behaviour of drivers and the
propagation of queues are explicitly represented. The connection to actual conditions
is guarantied by field measures at discrete points (loop detectors, speed radars, video
cameras, probe vehicles) that are collected in real-time and sent to a control room,
where sophisticated algorithms allow to reconstruct the current and future traffic pat-
tern on the entire network, hence extending the available data in time and space to
achieve a useful information for driver navigation and transport optimization.

1. Introduction

1.1 Overview of the technology

OPTIMA relies on a priori estimations of the traffic evolution during

each day-type that are accomplished through the simulation of the whole
transportation system, which reproduces the path choices of drivers travel-
ling on the congested road network from their origin to their destination at
specific instants. Then, the real-time measures are used to calibrate and cor-
rect such a mobility model, adjusting the base estimation to the current traf-
fic conditions, thus providing a very robust and reliable prediction.
Our approach, primarily based on the physical interpretation of the traffic
phenomena, differs substantially from the mere interpolation of field meas-
ures through artificial intelligence methods. Most monitoring systems apply,

1
Dipartimento di Idraulica Trasporti e Strade, Sapienza Università di Roma

1
2 Lorenzo Meschini, Guido Gentile

in fact, data mining techniques to match the current time-series with histori-
cal patterns, thus providing forecasts only on local conditions. However, sta-
tistical inference alone may not allow to deduce the traffic state of non moni-
tored links from the observed data or to figure out the consequences of un-
predictable and atypical events such as road accidents.
OPTIMA is specifically conceived for metropolitan contexts, where the
congestion is strongest, while the day-to-day variability and the within-day
fluctuation of vehicle flows and travel times is not negligible; but it can also
be applied in extra-urban frameworks, that are less complex by their nature.
The main advantage of OPTIMA, compared to the existing methodolo-
gies, is certainly the limited amount of real-time information that it needs to
reconstruct the traffic pattern. This feature makes this technology suitable to
set-up a traffic monitoring system also in smaller towns, where budget con-
straints are tight considering the number of potential users, or complemen-
tary to rapidly expand an existing system in terms of space (covered area)
and density (controlled links).
OPTIMA integrates traffic monitoring and prediction with route guidance
and vehicle routing, thus letting its comprehensive information system fully
available for many off-line and real-time applications concerning mobility
management and transport services.
The platform is constituted by three main components:
• OPTIMA Traffic, an efficient control system for real-time traffic
monitoring and short-term prediction of its evolution, based on the actual
dynamics of vehicle flows and queue propagations;
• OPTIMA Navigator, a powerful navigation system that continuously
updates the driver’s path toward its destination, based on the traffic condi-
tions that are estimated and predicted for the entire road network;
• OPTIMA Fleet, an easy management system for vehicle tracking on
the road graph and optimal routing, to be employed in public transport and
freight distribution.

1.2 Context and motivations

The mobility of goods and people is a critical issue for our society. Thus,
the ability to predict traffic conditions on the network and to make decisions
about the optimal trips of users and vehicles allows for a more efficient ac-
tivity planning by individuals and companies, achieving substantial savings
for the society as a whole.
 Real-time traffic monitoring and forecast through OPTIMA 3

To faced these needs, in recent years there has been a significant techno-
logical development of the so-called Intelligent Transportation Systems
(ITS). The diffusion of automatic systems for monitoring urban and extra
urban road networks, has made now available a vast amount of data (flow,
density and speed of vehicles on given sections), both historical and real-
time, describing the state of many links and their evolution over time.
In parallel, the current deployment of technologies for mobile communi-
cations and satellite positioning permits providing to individual travellers
real-time information to support dynamic decisions and adaptive strategies.
We consider in particular the possibility of suggesting to drivers optimal
paths with respect to actual and forecasted traffic conditions on the entire
network, and in particular to the estimations of travel times.
However, it is evident that the methodologies used so far for infomobility
are now totally inadequate to exploit the enormous potential of the ITS sec-
tor described above. Suffice it to say that the navigation systems, now in-
creasingly spreading on the market, provide indications based on static speed
limits, denying any explicit effect of road geometry and of traffic congestion.

1.3 Objectives and outputs

OPTIMA is conceived to bridge the gap between information widely

available but not sufficiently complete to be usable by the decision maker
(whether a supervisor or an individual), and a set of mobile communications
technologies that become mature to receive and transmit real-time informa-
tion to users.
The primary objective for OPTIMA is to make a decisive step forward in
providing traffic information and optimal routing that are more effective, re-
liable and useful.
In particular, OPTIMA is able to:
• infer the flow state (flow, speed and density, i.e. queue) occurring on
some links, from observations of traffic conditions carried out on the net-
work, especially in the form of vehicle trajectories;
• extend the measures made for a limited number of links by obtaining
an estimate of the full network status and of its evolution in the next future;
• retrieve, based on these state variables, an estimation of travel times
on each link of the network depending on the entering instant;
• identify, based on the travel times and costs on the network, an op-
timal path between any pair of nodes depending on the departure instant;
4 Lorenzo Meschini, Guido Gentile

• operate navigation and routing in real-time, that is continuously up-

dating the estimates of travel times and of optimal paths on the basis of the
data collected from the network.

1.4 Key features and advantages

Compared to the existing approaches and methodologies for the estima-

tion and prediction of road travel times, OPTIMA presents the following in-
novative features and key differences:
• it provides travel times that are not constant over time, but instead
are varying during the day depending on the estimated and/or observed traf-
fic congestion; these are definitively more accurate and reliable than those
actually provided by other technologies;
• based on the above information and on the time-varying characteris-
tics of the road network (i.e. limited access zones), it is able to perform real
time calculation of optimal paths depending on the departure time;
• both the off-line and the real-time estimation and prediction of travel
times do not require a massive or extensive collection of traffic data; on the
contrary, they just need a limited amount of observed data; in particular
about one moving (vehicle) or fixed (camera) probe for every km of moni-
tored road network: to make an example, the principal network of the Rome
municipality, Italy, that consists of about 1000 km of roads and serves about
3 Millions of citizens, requires about 1000 probe vehicles or traffic cameras;
the system is able to extend both in time and space these field measures by
means of a sophisticated methodology, integrating them with historical mo-
bility data;
• traffic information are collected, transmitted and elaborated auto-
matically, without the need of human operators; this results in a dramatic re-
duction of the time needed to make traffic information available for mobility
decisions (down to few minutes);
• it is scalable and replicable, since it is base on commercial standards
or publicly available data regarding transportation mobility and socio-
economical activities; moreover, it doesn’t rely on external information
sources prided by municipalities, police, or transport network operators (al-
though it can use these data, if available), since observations may come from
the same vehicles that use the system and act as automatic moving probes.
 Real-time traffic monitoring and forecast through OPTIMA 5

2. Functional description of the platform

OPTIMA Traffic is a software system for monitoring traffic in real-time

and forecasting its short-term evolution; it constitutes the beating heart of the
ITS platform and represents its more innovative module from a methodo-
logical point of view.
OPTIMA Traffic is indeed based on advanced models and algorithms for
the dynamic assignment of transport demand on road networks, that allow to
estimate travel times, traffic flows and vehicle queues in their daily variation
during the day, even starting from an extremely contained number of obser-
vations about the congestion state, with respect to other existing methodolo-
gies that are based on statistical inference of field measures.
OPTIMA Traffic is a modular system based on a client-server architec-
ture that is founded mainly on four software components; the first two off-
line and the second two on-line.
OPTIMA Traffic is based on a methodological framework whose func-
tional structure, depicted in the diagram of Figure 1, is briefly described be-
low by identifying the modelling components (rounded gray), the input and
output (rectangular blank) and their relationships (arrows).

The base model of the transportation system is build starting from Census
and network data by means of semi-automatic procedures stored within the
TMB.
The base model is firstly used off-line to calculate an a priori estimation
of travel times and path choices applying the DUE algorithm. The latter are
represented by turn splitting rates.
Real-time data on the state of the road network are collected by moving
probes (equipped vehicles) and/or fixed probes (e.g. loop detectors), in the
form of space-time trajectories or flows and speeds, respectively. The Vehi-
cles Tracker module transforms the observed trajectories into traffic meas-
ures of speeds and capacities on the monitored links of the road graph.
The SDNL software is responsible for putting together the real-time ob-
served data with the simulation providing the a priori estimation. This is
done by correcting in rolling horizon the resulting vehicle queues while tak-
ing into account the base path choices of drivers to reproduce as closely as
possible the actual traffic conditions on the network. The calibrated simula-
tion constitutes then the forecast of travel times that can be used in the route
guidance module to provide optimal paths.
6 Lorenzo Meschini, Guido Gentile

CENSUS DATA NETWORK DATA PROBES

(inhabitants, workers) (street map) (cars, loops, cameras)

TRANSPORTATION VEHICLE
MODEL BUILDER TRACKER

SUPPLY AND DEMAND TRAFFIC MEASURES

(graph, OD matrices) (speeds, capacities)

DYNAMIC USER
EQUILIBRIUM

BASE TURN
SPLITTING RATES

SEQUENTIAL DYN
NETWORK LOADING

BASE ESTIMATION REAL-TIME FORECAST

OF TRAVEL TIMES OF TRAVEL TIMES

ROUTE
GUIDANCE

OPTIMAL
PATHS
Figure 1. OPTIMA Traffic functional diagram.

2.1 TMB - Transportation Model Builder

This software integrates into a GIS graphical interface a methodology

that allows to determine, through a chain of models and algorithms, a trans-
port graph, a territorial zoning and origin-destination demand matrices, star-
ing from commercial street networks (e.g. Navteq, TeleAtlas) and standard
socio-economical data (e.g. ISTAT, ISFORT, for Italy) in a totally automatic
manner. Thus TMB allows the creation of truly predictive traffic simulation
for large urban areas with few days of calibration by relying mostly on pub-
 Real-time traffic monitoring and forecast through OPTIMA 7

lically available data. This tool lets to set-up reliable data bases for traffic as-
signment algorithms with low costs for planning studies aimed at mobility
management and for monitoring systems aimed at navigation/routing.

2.2 DUE - Dynamic User Equilibrium

This software solves in suitable computing time the Dynamic Traffic As-
signment to large road networks, simulating the formation, the dispersion
and the retro-propagation (spillback) of queues during the day both in met-
ropolitan and extra-urban areas. For a given time-varying demand of trips
from origins to destinations, DUE provides the within-day trend of car flows,
vehicle queues, travel times and polluting emissions for each road link at
equilibrium, together with the generalized costs of each user and the corre-
sponding trip choices, in particular his trajectory on the network.
DUE allows to analyze multiuser and multimodal contexts with elastic
demand, including the departure time choice. It can calculate the dynamic
equilibrium (consistency among, costs, times and flows within the path
choices) or simply the dynamic network loading (traffic mechanics for given
path choices). A module for the adjustment of o-d matrices and their daily
evolution from traffic counts is also available.
DUE is the first software that can handle detailed metropolitan networks,
simulating a whole day in real-time. This result is possible because DUE
manages relations among entire temporal profiles, allowing thus to consider
long time intervals (minutes), contrary to micro and meso simulators that re-
quire short time intervals (seconds).

2.3 Vehicle Tracker

This client-server application is designed to track on the road graph the

trajectory of a vehicle that carries on-board a mobile device (e.g. a cellar
phone) equipped with a satellite positioning system (e.g. GPS and in the next
future Galileo) and with an internet connection (GPRS or UMTS). The client
side records the time sequence of the space positions (geographic coordi-
nates) followed by the mobile device and sends these data to a server at
regular intervals through the wireless connection. The server side collects the
packets of space-time points sent by the mobile device and project them on
the road graph, filtering out the data that imply unrealistic trajectories
8 Lorenzo Meschini, Guido Gentile

through a graph matching algorithm, so as to reconstruct the path actually

followed by the vehicle and determine the trend of its speed.

2.3 SDNL – Sequential Dynamic Network Loading

This software is designed to provide real-time estimates and short-time

forecasts of congestion conditions on all road links, starting from traffic data
(speeds and flows) continuously retrieved at discrete points on the network
from moving probes (equipped vehicles) and/or fixed probes (video cameras,
speed radars, loop detectors). The underlying mathematical model is based
on the explicit representation of mobility on the network as a dynamic equi-
librium between transport demand and supply, that simulates, on one side,
the vehicle trips of and the path choices, and on the other side, the dynamic
of flows and the propagation of queues, using the General Link Transmis-
sion Model.
SDNL aims, therefore, at reproducing the traffic phenomena following a
deductive approach, rather than an inductive one, as done instead by the ex-
pert systems that are often utilized for infomobility applications. The advan-
tage is to obtain a comprehensive picture of the traffic pattern and of its
short-term evolution, even under conditions never experimented before such
as road works or accidents, by projecting the available point measures of
speed and flows in space and time. But to this end, SDNL needs more in-
formation on the road network features and on the transport demand charac-
teristics; in particular, it requires to pre-determine for the different day types
and the different hours the splitting rates at nodes, that can be obtained using
TMB - Transportation Model Builder and DUE - Dynamic User Equilib-
rium.
SDNL operates in rolling horizon: the current measures of flow and speed
are transformed in vehicular densities on the road links that periodically cor-
rect the number of cars predicted by the model; then, the process that simu-
lates the propagation of flows and queues is restarted. Differently to flows,
queues have a certain persistency; consequently, the effect of a correction
goes way beyond the travel time to reach the destination employed by the
vehicles that generated it. The current forecast is then influenced by all the
measures arrived so far. Moreover, the corrections propagate following the
node splitting rates, and not the vehicles that generated them, spreading onto
the network and updating the travel times of non monitored links.
 Real-time traffic monitoring and forecast through OPTIMA 9

References

We list below only a few references to our own works that present the
models used in the platform and can be used as an entry point to the really
vast existing literature on dynamic traffic assignment and route guidance.

Gentile G., Meschini L., Papola N. (2004) Fast heuristics for continuous dynamic
shortest paths and all-or-nothing assignment, presented at AIRO 2004, Lecce, Italy
Gentile G., Meschini L., Papola N. (2007) Spillback congestion in dynamic traffic
assignment: a macroscopic flow model with time-varying bottlenecks, Transpor-
tation Research B 41, 1114-1138 - rank n.2 in 2007 TRB papers
Gentile G. (2008) The General Link Transmission Model for dynamic network load-
ing and a comparison with the DUE algorithm, in Proceedings of the Second In-
ternational Symposium on Dynamic Traffic Assignment – DTA 2008, Leuven,
Belgium - nomination as best paper of the conference